The rise of generative AI presents a dilemma due to the public’s pragmatic acceptance of AI-generated design works (M=5.33, SD=1.89), alluding to the possibility of creative labor displacement. Grounded in Christensen’s Innovator’s Dilemma and Mori’s Uncanny Valley, this study examines how the Indonesian public perceives the ethical and utilitarian tensions of AI adoption. Using a sequential explanatory mixed-methods approach, an online survey (n=553) was conducted with respondents aged 20 to 50 in 10 Indonesian cities. Participants evaluated four case studies—advertisement, book cover, Instagram post, and photo manipulation—alongside their general sentiments. Findings indicate lower acceptance of GenAI for commercial (M=4.78, SD=1.84) than for personal use (M=5.43, SD=1.58), and concerns about GenAI’s potential to replace designers (M=5.2, SD=1.70). The lowest receptivity was observed in video and photo manipulation, reflecting the uncanny valley effect. Meanwhile, respondents tend to justify the use of GenAI when there are no formal regulations, thereby diminishing their ethical concerns, while also exhibiting difficulties in identifying AI-generated images. These perceptions underscore the importance of AI governance in protecting human designers from being replaced by machines and ensuring the authenticity of design works.

ABSTRACT As artificial faces and deepfakes become more prevalent, distinguishing between humans and artificial beings is crucial. Previous studies show that when a face transitions from clearly artificial to clearly human, ratings of affinity increase until the face becomes almost human. Here, the ratings will suddenly drop and quickly recover, revealing the uncanny valley. The faces in the uncanny valley elicit aversive reactions due to their subjective eeriness, and we investigated whether eeriness results in any special treatment in visual processing in humansby conducting a series of cueless temporal order judgment (TOJ) experiments. In Experiment 1, we confirmed the stimulus set. In the following experiments, participants identified which of the two faces appeared first, with faces having equal levels of eeriness (Experiments 2a and 2b) or different levels of eeriness (Experiments 3 and 4). We conclude that eeriness may not be the sole driver of attentional prioritization with uncanny valley faces.

Background: Facelift and blepharoplasty outcomes vary due to numerous factors. While patients once relied on before-and-after photos, they now increasingly look to AI-generated images for postoperative expectations. Due to the dataset training limitations of these AI models and the risk of unrealistic patient expectations, we sought to evaluate these AI models. Material and Methods: We utilized AI platforms DALLE, GetIMG, and Perchance to generate pre and post-operative images for facelift and blepharoplasty patients. A group of board-certified plastic surgeons and plastic surgery residents evaluated these images using 11 criteria, divided into categories of realism and clinical value. ANOVA and Tukey HSD Post-hoc statistical tests were used for data analysis. Results: Realism and clinical value showed no significant differences across AI models ( P = .21 and P = .59, respectively), yet GetIMG, (Mean ± Standard Deviation), (3.1 ± 1.12) significantly excelled in texture mapping against the others ( P < .01 and P = .03), and surpassed DALLE in age simulation accuracy ( P < .01). Across all models, healing and scarring prediction was the lowest performing metric (1.70 ± 0.42 P < .05). The evaluators also underscored the “uncanny valley” phenomenon. Conclusion: No significant differences were observed among the models’ realism or clinical value. Improving AI with real patient pre- and post-surgery photos is important to enhance accuracy and usefulness for surgeons and patients. Future research aims to compare AI-produced images with actual surgical photos and broaden the pool of expert evaluators.

Based on the uncanny valley theory, this study uses the scenario experiment method to explore user acceptance willingness toward AI-generated advertisements and its influence mechanism, and analyzes the roles of perceived eeriness, perceived trust, and perceived effort in this process. The study finds that perceived eeriness has a significant negative impact on user acceptance willingness toward AI-generated advertisements, while perceived trust and perceived effort play a mediating role between perceived eeriness and user acceptance willingness. This study provides a theoretical basis and practical implications for advertising practitioners and technology developers to optimize advertising design.

This paper investigates the psychoacoustic responses of unease produced by two popular techniques of sound synthesis in the modern age: granular synthesis and physical modeling synthesis. Both methods have the capability of generating unsettling sounds, which are discussed and analyzed based on their techniques and applications. Granular synthesis evokes discomfort through randomness, overlapping grains, and roughness. In contrast, physical modeling synthesis unsettles listeners through hyper-realistic instrument simulations, triggering the uncanny valley of sound. Comparing these two techniques, this study highlights the unique ways in which each synthesis method influences human auditory systems and findings suggest that physical modeling synthesis may provoke a strong sense of unease, offering valuable implications for sound design choices in music in the future.

The article presents a comprehensive theoretical and analytical review of the psychology of social perception of artificial intelligence (AI). The main aim of the study is to systematize the psychological foundations of how humans perceive AI not merely as a technological tool but increasingly as a social actor. The paper outlines the classical theoretical approaches to social perception (attribution theory, stereotyping, anthropomorphism) and demonstrates their relevance for the interaction “human – technological agent.” Particular attention is paid to attribution processes, projection and transference, anthropomorphism, the phenomenon of trust, and the “uncanny valley” effect. The article integrates findings from recent empirical studies that reveal the ambivalence of human attitudes toward AI. While AI is often perceived as useful and competent in functional contexts (education, healthcare, professional assistance), it is regarded as less capable in moral and ethical judgments, which significantly decreases trust in sensitive domains. Case studies with platforms such as ChatGPT, Replika, Siri, Copilot, and other virtual assistants illustrate how users tend to establish “quasi-relationships,” attribute emotions or intentions to non-human agents, and gradually normalize their interaction through mechanisms of social cognition. Cultural and media narratives play a crucial role in shaping social perception. In Western societies, dystopian frames such as “loss of control” or “Pandora’s box” prevail, while in East Asian cultures AI is more often represented within the narrative of “social progress.” For Ukraine, the perception of AI remains at an early stage: survey data demonstrate predominantly superficial awareness and low readiness for use, accompanied by distrust and insufficient understanding of ethical risks. The study concludes that the social perception of AI is a multidimensional phenomenon combining cognitive, emotional, and cultural mechanisms. It emphasizes the need for further research in Eastern Europe, where systematic psychological studies are still lacking.

As real-time live-streaming commerce has increasing applications of AI avatars, identity disclosure emerges as an important factor in shaping consumers short-term trust. An analytical model was constructed, integrating multimodal trust cues and causality inference to investigate the nonlinear threshold effects of AI identity disclosure. Based on large-scale real data and machine learning methods, the research establishes an optimal disclosure interval between 0.32 and 0.52, which significantly enhances trust while avoiding the trust decline caused by the uncanny valley effect. Depending on the age, different types of responses can be distinguished. Younger users have a lower tolerance for high-intensity information disclosure, while the middle-aged and elderly groups tend to prefer a moderate level of information disclosure intensity. Through double machine learning based on mediation analysis, parasocial intimacy and perceived truthfulness are established as prominent psychological mechanisms in cultivating trust. These findings provide a theoretical foundation for disclosing in a personalized form relative to distinctive user groups and provide practical expertise for designing trustworthy AI systems as well as for informing policymaking in governing such technologies.

The mainstream launch of generative AI video platforms represents a major change to the socio-technical system of digital media, raising critical questions about public perception and societal impact. While research has explored isolated technical or ethical facets, a holistic understanding of the user experience of AI-generated videos—as an interrelated set of perceptions, emotions, and behaviors—remains underdeveloped. This study addresses this gap by conceptualizing public discourse as a complex system of interconnected themes. We apply a mixed-methods approach that combines quantitative LDA topic modeling with qualitative interpretation to analyze 11,418 YouTube comments reacting to AI-generated videos. The study’s primary contribution is the development of a novel, three-tiered framework that models user experience. This framework organizes 15 empirically derived topics into three interdependent layers: (1) Socio-Technical Systems and Platforms (the enabling infrastructure), (2) AI-Generated Content and Esthetics (the direct user-artifact interaction), and (3) Societal and Ethical Implications (the emergent macro-level consequences). Interpreting this systemic structure through the lens of the ABC model of attitudes, our analysis reveals the distinct Affective (e.g., the “uncanny valley”), Behavioral (e.g., memetic participation), and Cognitive (e.g., epistemic anxiety) dimensions that constitute the major elements of user experience. This empirically grounded model provides a holistic map of public discourse, offering actionable insights for managing the complex interplay between technological innovation and societal adaptation within this evolving digital system.

This study investigates how Chinese undergraduate music students’ perceptions of AI-generated content (AIGC) are affected by generative artificial intelligence (GAI). To explain students’ acceptance of generative AI, the study integrates the Stimulus–Organism–Response (SOR) framework with the Technology Acceptance Model (TAM) and the Unified Theory of Acceptance and Use of Technology (UTAUT). A mixed-method approach was employed, involving 600 university students through quantitative surveys and qualitative interviews. The analysis explores students’ responses to the uncanny valley effect, perceived usefulness and ease of use of anthropomorphic features, and intention to adopt the technology. Findings indicate that although the human-like aspects of AIGC cause discomfort, quantitative data show that students find components like voice interaction and emotional expression helpful for learning music. Qualitative evidence further reveals adaptive strategies to mitigate discomfort, including integrating AIGC with peer review. The study concludes that AIGC holds significant potential for enhancing music education but underscores the need to address the uncanny valley effect to foster greater emotional engagement. To better accommodate diverse student needs, future research should investigate potential long-term effects and support the development of customized AIGC tools.

The article presents experimental verification of the possibility of achieving the level of entry into conversational practice in German from zero level exclusively through interaction with AI-based tutors within the framework of the author’s hyperintensive method of language learning. The author conducted a two-month (62 days) self-experiment from December 2024 to January 2025. During the experiment, the hyperintensive method was applied, adapted for conducting lessons with personalized AI tutors based on the Student Operated Lesson approach. As part of the experiment, an ecosystem of 21 specialized AI tutors was developed on the ChatGPT Plus, Google AI Studio, Claude, and ElevenLabs platforms, divided into two functional groups: academic tutors for maintaining complex educational dialogues, and conversational tutors who acted as language partners for informal communication. Throughout the experiment, the progress of conversational skills was monitored with the involvement of independent experts (teachers of German). The author conducted a quantitative linguistic analysis of transcribed dialogues using specially developed software based on the spaCy and Textstat libraries. Over the 62 days of the experiment, with a total time expenditure of 131 hours, of which 52.5 hours were devoted to targeted conversational practice with AI agents, the author achieved the level of entry into conversational practice. The results of the experiment were verified through final spontaneous conversations with six unfamiliar German speakers and two independent experts, who determined the level achieved to be B1 on the CEFR scale in terms of conversational skills. Linguistic analysis revealed the acquisition and use of 1,060 unique lemmas in spontaneous speech, progressive complication of syntactic structures (increase in dependency depth), a decrease in the number of hesitation markers, which characterizes an increase in fluency, and a decrease in readability, which indirectly indicates a complication of speech structures. The level achieved demonstrated stability during a follow-up check three months later. Specific psycholinguistic features of interaction with AI were identified, including the “uncanny valley” effect and the need for high self-discipline on the part of the learner. The experiment confirms the applicability of the hyperintensive method to learning conditions exclusively with AI tutors and the possibility of achieving functional fluency in foreign languages without the participation of human teachers. The results open up new prospects for the development of both autonomous language learning methods and hybrid language learning models that integrate the advantages of AI technologies and traditional pedagogical approaches to foreign language learning. Nevertheless, further research should focus on studying the long-term sustainability of the results achieved and optimizing human interaction with AI systems in the educational process.

Authenticity, ethics, and transparency in virtual influencer marketing: A cross-cultural analysis of consumer trust and engagement: A systematic literature review.

Workplace bullying is closely related to poor work states. Previous studies have primarily explored the binary relationship between perpetrators and victims, with limited research examining the emotional exhaustion of bullying roles from the perspectives of victims and bystanders. Therefore, this study recruited 597 participants and conducted a scenario-based experiment to investigate whether generative AI can alleviate the poor work states of bullying roles in the medical workplace, thereby demonstrating the interaction between generative AI’s information delivery methods and bullying roles in relation to emotional exhaustion. The results showed that bullying roles in the medical workplace significantly influence emotional exhaustion, with victims experiencing significantly higher levels than bystanders. Moreover, generative AI’s information delivery methods can effectively moderate the work states of victims. Thus, this study advances the field of human-computer interaction by shifting its focus from functional adaptation to emotional ecology. It also provides empirical evidence from medical scenarios for the uncanny valley theory. Furthermore, this research lays a theoretical foundation for the design of emotional interaction functions in medical AI systems.

From approach to avoidance: How AI agent cognitive and affective empathy elicits the uncanny valley effect

ABSTRACT E-commerce businesses utilise automated text-based dialogue systems (chatbots) to provide pre and post-sale assistance. However, due to technical limitations and repetitive responses, these chatbots often fail to engage spontaneously with customers, resulting in consumer frustration and subsequent brand switching. Therefore, this research investigates the impact of eeriness on consumers’ purchase intention in the presence of human likeness (i.e., appearance, skills, and style). Scenario-based survey was conducted with 355 participants to explore the eeriness impacts on consumers’ purchase intentions based on the uncanny valley effects theory. The hypotheses were tested using the partial least squares method. Findings reveal that chatbots with a cartoonish appearance increase consumers’ purchase intention by reducing the eerie effect. Consequently, findings also suggest that tailored skills and a warm style may mitigate eeriness and enhance PI under certain conditions, compared to response variety and competent styles. This study extends the Uncanny Valley Theory (UVT) by introducing the concept of ‘botmorphosis’, highlighting how shifts in chatbot appearance, skill, and style jointly shape user perceptions and behaviour in e-commerce contexts. The findings of this study provide valuable insights for developers and digital businesses keen to use chatbots as customer assistants in e-commerce settings to improve the interaction between human and AI.

The uncanny valley effect describes a phenomenon where humanoid, almost lifelike virtual agents evoke feelings of discomfort in the observers. The Pathogen Avoidance Hypothesis proposes that these feelings are based on a cognitive mechanism that originally evolved to motivate pathogen avoidance. Slight imperfections in virtual agents are thereby misconceived as cues of communicable diseases that elicit disgust and avoidance behavior. Whether uncanny entities also increase mucosal immune responses, particularly when immediate avoidance is impossible, currently remains elusive. The present study examined the link between the uncanny valley and mucosal immune responses by comparing changes in the salivary antibody secretory immunoglobulin A (sIgA). For this, we developed three sets of virtual agents (cartoon, uncanny, and realistic agents) that differed in their human-likeness and uncanniness. Agents were encountered in a social virtual reality (VR) task that required close approach, making avoidance impossible. We found that only the set of uncanny agents who showed slight deviations from normal human appearance, yet were still considered as more human-like than the cartoon agents, evoked a significant increase in sIgA release during social interactions in VR. This conforms to error-management theories suggesting that perceptual systems are biased towards false positive responses to potential contagion cues, reducing the risk of missing actual health threats. Our findings further indicate that presence in virtual reality was likely essential for effective immune activation by uncanny agents, as increased sIgA levels correlated with deeper involvement in the virtual environment. Moreover, the immune response may have been automatically triggered by perceived threats from these agents, as conscious evaluations of interoceptive awareness, state anxiety, and eeriness during VR did not correlate with higher antibody release. Altogether, these data show that perception of uncanniness in VR evokes mucosal immune responses resembling those related to real contagion threats.

ABSTRACT Recent advancements in Generative Artificial Intelligence (GenAI) have demonstrated its capability to produce creative outputs that closely resemble human creations, particularly in image generation. This technological leap presents significant opportunities and challenges for educational research and practice. To better understand student perceptions of GenAI, this study utilized deep learning-based text mining techniques to analyze 125,952 Weibo posts. The analysis identified six key themes, including application scenarios, personalized expression, technical modeling, prompt engineering, attitudes of educators and institutions, and legal issues. Sentiment analysis revealed that while positive sentiments prevail, negative sentiments fluctuate periodically. Co-occurrence network analysis highlights the sources of negative sentiments, including concerns about spurious images, perceived complexity, employment anxiety, privacy and ethical concerns, and the uncanny valley effect. These findings offer actionable insights for educators, researchers, and policymakers to harness GenAI opportunities.

ABSTRACTAs artificial intelligence (AI) technologies progress, AI agents arise as potential teammates in the workplace. This study explores how the visual representation of the AI agent as well as its conformity to traditional gender stereotypes affects the manifestation of uncanny valley effects in a workplace team context. Using social categorization theory, we conducted three between‐subject, randomly assigned experimental studies (N1 = 239, N2 = 513, N3 = 403, NTotal = 1155) and investigated how the AI agents' varying levels of human‐likeness (human‐like, almost human‐like, or cartoon‐like) and conformity to traditional gender stereotypes (gender‐neutral, male, or female versus warm or competent) in two different workplace contexts (Study 2: traditionally masculine, Study 3: traditionally feminine) affect eeriness reactions, initial trust, and willingness to have the AI agent join the all‐human team. In Study 1, we find that a more human‐like visual representation is associated with lower eeriness reactions, higher initial trust, and higher willingness to have the AI agent join the team. Once gender and temperament are introduced in Study 2, the warm male AI agent leads to higher eeriness reactions by not conforming to gender stereotypes while the warm female agent elicits the lowest eeriness reactions. When the workplace context is changed into a more traditionally feminine setting in Study 3, the warm male agent elicits the lowest eeriness reaction despite not conforming to gender stereotypes.

In this study, we examined the impact of agent familiarity and knowledgeability on several variables spanning agent perceptions (i.e., perceived knowledge, familiarity, trust, anthropomorphism, uncanny valley effect, and likability), social and emotional experiences (i.e., co-presence, rapport, cognitive process expectations, and willingness for future interaction), and conversation dynamics (i.e., conversation transcript, participants’ response word count, and response time). We created two virtual agents for the study: a digital replica of a professor from our department (i.e., familiar agent) and an agent with similar demographic variables (i.e., age, gender, and ethnicity) but with a fabricated appearance and voice (i.e., unfamiliar agent). We implemented both agents to exhibit two levels of knowledgeability (i.e., low and high) in the domain of game development and course-specific information. We used large language models (LLMs) to provide the agents with persona information and domain knowledge through prompt engineering. For our user study, we followed a 2 (familiarity: unfamiliar vs. familiar agent) \(\times\) 2 (knowledgeability: low vs. high knowledgeability) within-group study design and recruited 32 participants who engaged in a five-minute, conversation-based virtual reality (VR) interaction with all four experimental conditions: unfamiliar agent with low knowledgeability (ULK), unfamiliar agent with high knowledgeability (UHK), familiar agent with low knowledgeability (FLK), and familiar agent with high knowledgeability (FHK). The findings demonstrated a significant main effect of agent familiarity on perceived knowledge, suggesting that familiarity plays a crucial role in shaping users’ perception of the agent's knowledgeability level. Besides perceived knowledge, familiarity also affected all other variables, apart from co-presence. Conversely, agent knowledgeability affected perceived familiarity, trust, anthropomorphism, cognitive process expectations, willingness for future interaction, conversation content, and participants’ response word count. Finally, we found an interaction effect between agent familiarity and perceived knowledge, indicating that familiarity has a significant influence on users’ perceptions of the agent's knowledgeability. This study contributes to the field of conversational human-agent interaction in VR by providing empirical evidence on how adapting both familiarity and knowledgeability of virtual agents can significantly enhance user experience, offering valuable insights into designing more engaging, trustworthy, and effective embodied conversational agents.

Virtual Digital Human Pedagogical Agents (VDHPAs) have emerged as vital tools in online and blended learning environments. However, the extent to which they improve learning outcomes in Virtual Reality (VR) environments remains inconclusive. This meta-analysis synthesizes findings from 36 empirical studies conducted between 2013 and 2023 to examine the effectiveness of VDHPAs in VR-based learning. Specifically, we analysed two process-oriented variables—cognitive load and social presence—and three outcome-oriented variables—retention, transfer, and other assessment types. The results indicate that while VDHPAs do not significantly reduce cognitive load (g = -0.084), they significantly enhance learners’ social presence (g = 0.402). Additionally, VDHPAs were found to improve retention (g = 0.451), transfer (g = 0.288), and other test scores (g = 0.423). Moderator analyses revealed that the effects vary depending on agent design features (e.g., gestures, voice, facial expression), content characteristics (e.g., subject domain, knowledge type), and learner attributes (e.g., education level, prior knowledge). This review further discusses the implications of agent embodiment, the "uncanny valley" in affective response, and challenges in long-term outcome assessments. The study contributes to a deeper understanding of how to optimise VDHPAs for immersive learning experiences and highlights directions for future interdisciplinary research in educational technology and digital arts.

This paper develops the uncanny ridge as a thought experiment within debates on AI alignment, intersystemic communication, and the limits of novelty in machine learning. Repositioning systemic misalignment in AI as a generative rather than disruptive force, it draws on and inverts Masahiro Mori’s uncanny valley, which describes human discomfort toward near-human simulations. The uncanny ridge shifts focus from breakdowns in recognition to the conditions under which misrecognition between AI systems produces novelty. While alignment is often assumed to improve coherence, excessive synchronization leads to hyper-convergence, constraining AI’s capacity for emergent intelligence. To formalize this, the paper introduces the uncanny index, a metric for identifying when systemic misrecognition produces innovation rather than collapse. Case studies of mode collapse in generative adversarial networks (GANs), reinforcement learning with human feedback (RLHF) constraints, and multimodal AI architectures demonstrate that novelty does not emerge from complexity alone but from structured divergence and delayed alignment. Rather than treating misalignment as a failure, this thought experiment reframes the uncanny ridge as a dynamic site of intersystemic tension, where AI models engage in productive misrecognition, generating new computational strategies and unexpected modes of coordination.